Solana's new consensus, Alpenglow, just dropped and it's interesting!
Meet Votor—the voting engine that challenges traditional aspect of blockchain consensus.
Is it genius or dangerous?
Let's dive in 🧵 👇
1/11
Votor runs two voting modes simultaneously:
→ Fast track with 80% stake voting = 1-round finality
→ Safe track with 60% stake voting = 2-rounds finality
So, the blocks finalize in min(δ₈₀%, 2δ₆₀%)—whichever finishes first.
2/11
Here's how it works in practice:
→ Leader proposes block via Rotor (fast data dispersal)
→ Validators vote with notarization votes
→ If 80% agree, the block is finalized immediately
→ If 60% agree, there's a fallback round and then finalization happens
3/11
Votor uses 5 vote types:
→ NotarVote = Yes
→ NotarFallbackVote = Backup Yes
→ SkipVote = No
→ SkipFallbackVote = Backup No
→ FinalVote = Finalize
4/11
In the end, certificate thresholds are the key:
→ (Round 1) 80% notarization = Fast-Finalization
→ (Round 1) 60% notarization = Notarization Certificate
→ (Round 2) 60% finalization = Slow-Finalization
→ (Fallback) 60% fallback = Skip/Notar-Fallback
5/11
There are also safety checks baked into the protocol logic:
→ SafeToNotar = Prevents conflicting votes
→ SafeToSkip = Handles broken/lost blocks
&
→ Exclusive voting = Can’t vote for and against a same block
→ Parent validation = Ensures valid block chains
6/11
But here's where it gets interesting...
Traditional BFT requires 67% honest nodes (33% fault tolerance).
Alpenglow only requires 60% honest nodes with a new "20+20" failure model, assuming:
→ Up to 20% malicious actors
→ Up to 20% nodes offline due to crashes/outages
7/11
The rationale here is that in large-scale PoS systems with thousands of validators, coordinated 33% attacks requiring billions in stake are unrealistic. Most failures are crashes, bugs, misconfigurations—not coordinated attacks.
8/11
So, the gamble...
Why assume the worst-case scenario when you can optimize for the common case?
If it holds:
→ Faster finality
→ Lower latency
→ Better bandwidth usage
→ More real-world resilience
9/11
But, there are risks:
→ Deviate from 40+ years of 33%-tolerant BFT theory
→ Assume failures are usually benign
What if assumptions break? What if new attacks arise?
There is a tradeoff between proven security and performance.
10/11